Pediatric thyroid nodules: ultrasonographic characteristics and inter-observer variability in prediction of malignancy

Background: Pediatric thyroid nodules, while uncommon, have high malignancy risk. The objectives of the study were (1) to identify sonographic features predictive of malignancy; (2) to create a prediction model; and (3) to assess inter-observer agreement among radiologists. Methods: All available cases of thyroid nodules, surgically removed between 2000 and 2009. Three radiologists reviewed the sonographic images; 2 pathologists reviewed the tissue specimens. Adult prediction models were applied. Interobserver variability was assessed. Results: Twenty-seven subjects, mean age 13.1 +/- 3.4 years, were included. Nineteen nodules were differentiated thyroid carcinomas. On multivariate analysis, size was the only significant predictor of malignancy. On recursive partitioning analysis, size > 35 mm with microcalcification and ill-defined margins yielded the best prediction model. Radiologist inter-observer agreement regarding malignancy was moderate (kappa = 0.50). Conclusions: Larger size, microcalcifications and ill-defined margins on ultrasound demonstrate the best predictive model for malignancy in the pediatric population. Experienced pediatric radiologists demonstrate moderate inter-observer agreement in prediction of malignancy.